modules/yolov8_train.sh

#!/bin/bash

SCRIPT_DIR=$(cd $(dirname $0); pwd)
# Parameters
YOLO_DATASET_DIR=$SCRIPT_DIR/../export_yolo/YOLODataset
MODEL="yolov8n.pt"  # YOLOv8 model type
EPOCHS=1000
SAVE_PERIOD=50
OUTPUT_DIR=$SCRIPT_DIR/../export_yolo/train_output
USE_GPU_TUNING=true
RESUME_TRAINING=false

mkdir -p $OUTPUT_DIR/runs/tune

# Install ImageMagick if not already installed
if ! command -v identify &> /dev/null
then
    echo "ImageMagick not found. Installing..."
    sudo apt install -y imagemagick
fi

# Automatically detect image size from the training images
IMG_SIZE=$(identify -format "%wx%h\n" $YOLO_DATASET_DIR/images/train/*.png | head -n 1 | cut -d 'x' -f 1)

# If USE_GPU_TUNING is true, adjust parameters based on GPU memory
if [ "$USE_GPU_TUNING" = true ]; then
    # Get the GPU memory in MB
    GPU_MEMORY=$(nvidia-smi --query-gpu=memory.total --format=csv,noheader,nounits | head -n 1)

    # Set the batch size and learning rate based on GPU memory
    if [ "$GPU_MEMORY" -ge 24576 ]; then
        BATCH_SIZE=8   # For 24GB+ GPUs
        BATCH_SIZE_PER_IMAGE=512
    elif [ "$GPU_MEMORY" -ge 12288 ]; then
        BATCH_SIZE=4   # For 12GB+ GPUs
        BATCH_SIZE_PER_IMAGE=256
    else
        BATCH_SIZE=2   # For smaller GPUs
        BATCH_SIZE_PER_IMAGE=128
    fi

    # Adjust learning rate based on batch size
    BASE_LR=$(echo "0.00025 * $BATCH_SIZE / 2" | bc -l)
    echo "GPU memory: $GPU_MEMORY MB"
    echo "Batch size: $BATCH_SIZE, Batch size per image: $BATCH_SIZE_PER_IMAGE"
    echo "Base learning rate: $BASE_LR"
fi

cd $OUTPUT_DIR
# Run training with the YOLO model
python3 - <<END
from ultralytics import YOLO
import os
import ray
ray.init(ignore_reinit_error=True)

# Hyperparameters and settings
data_path = "$YOLO_DATASET_DIR/dataset.yaml"
model_path = "$MODEL"
img_size = $IMG_SIZE
batch_size = $BATCH_SIZE
epochs = $EPOCHS
save_period = $SAVE_PERIOD
output_dir = "$OUTPUT_DIR"
resume_training = $([ "$RESUME_TRAINING" = true ] && echo True || echo False)  # Convert bash boolean to Python boolean


# Load the model
model = YOLO(model_path)

# Automatically tune hyper-parameters with Ray Tune
print("Starting hyperparameter tuning...")
tune_results = model.tune(
    data=data_path,
    batch=batch_size,
    imgsz=img_size,
    augment=True,
    multi_scale=True,
    rect=True,
    scale=0.8
)
print("Tuning completed. Starting final training...")

# Start final training after tuning
results = model.train(
    data=data_path,
    epochs=epochs,
    save_period=save_period,
    patience=0,
    batch=batch_size,
    imgsz=img_size,
    project=output_dir,
    name="final_train_results",
    resume=resume_training,
    augment=True,
    multi_scale=True,
    rect=True,
    scale=0.8
)
print("YOLO training completed.")
END